next up previous
Next: Reverse Engineering Distributed Applications Up: Description of Research Projects Previous: Location Management in Cellular Systems

Parallel and Distributed Debugging

Our research group has been building a parallel and distributed debugger, named Poet. This tool provides process-time diagrams (or message sequence charts) of partially ordered parallel or distributed executions, where the time dimension is either based on the real time of event occurrence, or a model of logical time (the precedence relation introduced by Lamport) widely used in reasoning about causality in a distributed system. The novel features of Poet are its target-system independence (we use it to visualize the execution of applications written for DCE, PVM, SR, ABC++, C++, and others) and its unique abstraction features in both the process and the time domain, which allow for execution visualizations at higher abstraction levels. These features are described in [1, 8, 10, 13, 17, 26]. Poet also supports more traditional debugging operations such as single-stepping (see [9]), breakpoints, or deterministic replay. In addition, a traditional sequential debugger can be controlled through Poet for individual processes.

Thomas Kunz
Department of Computer Science
University of Waterloo
Waterloo, Ontario N2L 3G1
Fri May 2 10:11:04 EDT 1997